Before publishing a Midjourney image, answer three questions: Does it represent the subject honestly, do you have the rights to use it, and what disclosure does the publication channel require? Google's public guidance does not establish a blanket ranking penalty merely for using AI images. That does not make an invented product, customer site or test result reliable evidence.
Separate ordinary editorial images from Shopping product data
Google's generative-AI guidance emphasizes accuracy, quality and relevance and recommends giving readers useful context about automation. It separately points ecommerce publishers to Merchant Center's rules. These are not a universal requirement to place one particular watermark on every image on the web.
| Use | Check | What a caption can do |
|---|---|---|
| A conceptual blog illustration | Relevance and possible confusion with a photograph | Explain that it is an AI-assisted illustration and what it depicts |
| A product or performance visualization | Accuracy against the product actually sold | Identify a simulation, but not repair false specifications |
| An AI image submitted to Merchant Center | Applicable product-image policies and embedded IPTC metadata | Add context, without replacing file metadata |
| An image derived from another person's work | Permissions for both the input and intended output use | Credit a source where required, without granting permission itself |
Merchant Center's AI-content policy requires generative-AI images to contain IPTC DigitalSourceType TrainedAlgorithmicMedia metadata and says not to remove existing source tags. It also identifies synthetic-composite and algorithmic-media tags that should be preserved. The IPTC concept record gives the identifier http://cv.iptc.org/newscodes/digitalsourcetype/trainedAlgorithmicMedia. Use the type that describes how the asset was made; an arbitrary “AI” keyword is not equivalent.
Inspect the source file and then download and inspect the actual desktop and mobile variants delivered by the CDN. Compression or format conversion can change metadata. A tag in the local original does not prove that the submitted product image retains it. Removing provenance to evade recognition is not a sound publication workflow.
Does Midjourney require attribution for commercial use?
The Terms of Service effective May 27, 2026 state that users own assets they create to the fullest extent possible under applicable law, subject to the agreement and third-party rights. They do not make a uniform “Made with Midjourney” credit a general condition for every self-created commercial image. They do prohibit intentionally misleading recipients about an asset's nature or source.
- Business size matters: the terms require a Pro or Mega plan for a company, or an employee of a company, with more than US$1 million in annual revenue to own the assets. Do not assume the lowest paid plan covers every business.
- Upscaling another person's image does not transfer ownership: the original creator retains it. Obtain the relevant permission rather than crediting only the tool.
- Later cancellation does not automatically remove rights to assets already created: applicable law and third-party conditions still matter.
- Public generation is not exclusive delivery: the terms describe public viewing and remixing and the license granted to Midjourney. Paying for an account does not itself keep client material confidential.
The commercial-use help page also explains the business-revenue and other-creator exceptions. Stealth mode does not hide creations made in public Discord channels from people in those channels. Confidentiality, contractual permission and legal copyright protection are separate questions.
Give readers a useful disclosure
A contextual caption might say: “AI-assisted concept illustration showing a possible warehouse layout; not a photograph of a customer facility.” That is more informative than “Source: Internet”. Where the design includes imagined dimensions or features, distinguish it from the actual specification.
Captions explain origin and purpose to readers; alt text conveys necessary image information to people who cannot see it; embedded metadata supports systems that read it. They complement rather than replace one another. Do not stuff the alt attribute with tool names or invent details absent from the picture. The E-E-A-T and content-credibility guide is also relevant when deciding whether an illustration has been placed where a reader expects evidence.
Check the delivered asset, not only the prompt
- Inspect the picture: check text, hands, connectors, materials, brands and physical structure. Correct technical illustrations against real specifications.
- Record the source: retain generation records, tool and plan details, input permissions and any applicable client authorization. A record supports traceability; it does not itself create a copyright registration.
- Check the page: use discoverable image elements, meaningful alternative text, appropriate captions and correct dimensions.
- Check size and metadata: balance clarity with transfer cost and inspect converted files. There is no universal 80 KB or 100 KB ceiling for all useful images.
- Check the channel: editorial websites, Shopping feeds, ads and social platforms can have different requirements.
Google's image guidance covers discovery, responsive delivery and image quality, while the metadata reference explains source and licensing information. Supplying that information is not evidence of a ranking bonus.
When planning content around specific reader queries, decide whether the reader needs a product detail, an interface location or an explanation of structure before generating an image. Repeating similar pictures for keyword variants does not establish useful coverage or a traffic multiplier. If a page's performance changes after publication, use the AI-image delivery and traffic checks to separate content issues from loading faults.